AI researchers harness the power of IoT to prevent manufacturing failures

17
Image Credit: Piqsels

Tapping into the potential of IoT, two Turkish researchers have devised a “predictive maintenance system” that could prevent production downtimes from happening – and ultimately, alter the way we address manufacturing failures as a whole.

Assembly lines rely on accuracy and prediction to manufacture the objects that drive today’s world. 

That being said, a single fault can cause the entire production to collapse – leading to huge losses both for the manufacturer and its potential customers. 

Turkish researchers Serkan Ayvaz, an AI researcher from Bahcesehir University, and Koray Alpay from Procter & Gamble’s Gebze Development Center, are trying to prevent such huge losses with an innovative AI-driven system.

Advertisement

In their study published in Expert Systems with Application, Ayvaz and Alpay devised a data-driven predictive maintenance system for production lines in manufacturing.

By employing IoT sensors and AI learning methods, this innovative system detected potential failures before they even happened.

With an early detection system, operators can act with speed to drastically reduce manufacturing expenses that would have otherwise gone to costly, time-consuming repairs.  

The concept of IoT has seen usage in the manufacturing industry, but barriers still exist, including the lack of realistic predictions from real-time data.

“This is the first AI-based predictive maintenance system implemented in a real factory in this specific sector,” the researchers said.

Tackling the common industrial IoT pitfalls, Ayvaz and Alpay’s proposed system showed promising outcomes.

“The evaluation results show that our proposed predictive maintenance system is effective in capturing the signals of machinery failure using real-time sensor data, and it can help prevent potential production stops by taking preventative actions suggested by the system,” they reported.

Predicting the fall before it happens

While the production lines are made to operate flawlessly, they are still prone to malfunctions from time to time. These stops may occur as a result of equipment malfunction, operator errors, or environmental factors.

Unfortunately, whether they occur by chance or by mistake, a production stop’s effects are substantial. 

Predictive maintenance (PdM)  identifies these risks and predicts the next error before the actual failure takes place. 

Although relatively new, the maintenance technique has the potential to make manufacturing even more sustainable than it already is, while at the same time boosting overall productivity.

It is a highly effective system – if put into thorough use, that is,

One common issue with PdM is data flexibility, the system’s ability to process the data it receives to make predictions. 

Too much data, unlabelled data, algorithm biases, and even complex high dimensional IoT data are among the many data flexibility issues that challenge the PdM.

Using the right algorithm 

Despite the many provocations of data towards PdM, PdM can still significantly benefit from data collected from IoT devices. 

With the right AI infrastructure, data collected from these sensors can be streamlined right into the decision-making process – just how IoT ought to operate.

“With an integrated data streaming prediction model, a PdM system may be able to detect the failures, create alarms by applying specific rules, execute commands on the production systems, and send warning messages to the authorized officials in real-time.

The developed PdM in the study followed suit and used data collected from IoT sensors in real-time. Not only did Ayvaz and Alpay’s machine learning model evaluate changing conditions, it also estimated the best time to perform maintenance. 

The system works for the most frequent failure types. But the researchers are looking to expand in upcoming studies.

“For the future work, we plan to apply the system to other types of production lines in different settings,” they concluded.

Advertisement